Will an AI system design a pathogen which leads WHO to declare an emergency of some sort by the end of the July, 2030?
22
27
470
2030
22%
chance

Emergency could be an epidemic, a pandemic, or even smaller scale ones as long as WHO calls it an emergency or implies it to be an emergency

Get Ṁ200 play money
Sort by:

Does the pathogen have to be a bacteria or virus, or does this also include nanotech or other more exotic things?

predicts NO

Giving this the old Fermi estimation:

1) AI designs pathogen from scratch, unaided: 20%

2) AI designs pathogen with significantly input from human: 90%

3) Pathogen escapes laboratory: 10%

4) Pathogen spreads after escape to degree of causing an emergency: 50%

So depending on whether you’d accept 2 instead of just 1) we’re talking 1%-5%.

@Noit using Tool AI to at best speed up or merely support what humans are doing won't count

bought Ṁ50 of NO

Clarification: does this require the AI to have designed the pathogen unaided (scientist aka pathogen GPT “make me a virus”), or would a human who happened to use AI as part of their process trigger YES?

What do you mean by "an AI system designs"? Does the AI have to do it alone? What kind of human support is acceptable? Or is it enough that there is an AI system that supports the human designers in some way? To what degree does the design need to be supported?

@AlexbGoode the majority of heavy lifting has to be AI's work for the design part of the problem.

@AlexbGoode Hard to tell what is acceptable but easier to tell what is not acceptable. If I use chatGPT to format or beautify my (self-written) email to a friend, that doesn't count as chatGPT writing the email. If I tell chatGPT to write the email based on so and so information, and edit it myself a bit and send it, then that counts as chatGPT having written the email

@firstuserhere How do you define majority? What do you mean by heavy lifting? How does the process of creating a pathogen looks like? I am no biologist but I imagine a lot of the time spent is lab work and experiments. Since we don't have robots that can do this (yet) most of the heavy lifting in terms of experimentation has to be done by humans.
Is a yes resolution preconditioned on robots being able to do biology lab work?

@AlexbGoode AI iteratively guides the design and the experiments are done by the human researchers and fed back to the AI system for debugging and outputting a better design until it works

@AlexbGoode humans acting as agents for the AI is totally fine

predicts NO

@AlexbGoode Question specified “designed” so I had assumed any production was human. AI only providing instructions.

@Noit yes thats right

@firstuserhere Is any iterative optimization algorithm sufficient?

@AlexbGoode no, has to be "AI" or machine learning system